Harrison Chase, LangChain CEO and co-founder, takes the stage at Cisco Live! to discuss ambient agents.
Sabrina Ortiz/ZDNET
Until recently, AI solutions that can execute tasks on your behalf seemed futuristic. Now the era of AI agents is here, with nearly every company offering its own solution. On the horizon, though, is a more advanced and even more promising milestone — ambient agents.
On day three of the Cisco Live! conference, LangChain CEO and co-founder Harrison Chase took the stage to discuss ambient agents, a concept pioneered by his San Francisco-based company. As the name implies, these agents take cues from their environment to commit actions rather than waiting for human input.
Also: Why ads are coming to your favorite AI bots and you’ve only got yourself to blame
What are ambient agents (and the perks)?
Currently, the AI assistance that users receive is deterministic; that is, humans are expected to enter a command in order to receive an intended outcome. With ambient agents, there is a shift in how humans fundamentally interact with AI to get the desired outcomes they need; the AI assistants rely instead on environmental cues.
“Ambient agents we define as agents that are triggered by events, run in the background, but they are not completely autonomous,” said Chase.
He explains that ambient agents benefit employees by allowing them to expand their magnitude and scale themselves in ways they could not previously do. Rather than 1:1 interactions between human employees and agents, ambience enables up to millions of agents to run in the background simultaneously. Instead of being limited to the number of chat windows you can use, you can instead rely on the agents to initiate their own, in response to environmental cues.
It’s similar to the concept of ambient computing or ubiquitous computing, which blends computing power into our everyday lives in a way that is embedded into our surroundings to power experiences such as smart home ecosystems. Instead of individually turning on every power switch, an assistant can pick up on cues, such as the sun dimming, to execute a set of actions.
Ambient agents in the enterprise could accomplish a similar goal, with these AI assistants orchestrating workflows and even working with others to reach desired outcomes without human intervention to kick them off, unless it is of maximum importance. These AI agents combine human reasoning with the speed and intelligence of advanced AI models, unlocking new capabilities we have not been able to access before.
“There is actually a pretty powerful combination when you think about the kind of human empathy, human creativity, working together with the power and scale of AI,” said Nathan Jokel, SVP, corporate strategy and alliances, to ZDNET.
Overcoming hallucinations
When talking about these types of ambient agents with advanced capabilities, it’s easy to become concerned about trusting AI with your data and with executing actions of high importance. To tackle that concern, it is worth reiterating Chase’s definition of ambient agents — they’re “not completely autonomous.”
Chase makes this distinction because he stresses the need for a “humans-in-the-loop” approach for ambient agents to work effectively. A LangChain blog post identifies that humans will be necessary for ambient agents in a pattern of notify, question, and review.
Also: Your data’s probably not ready for AI – here’s how to make it trustworthy
- Notify — The agent alerts the human about an event of importance.
- Question — The agent asks a human for any direction or clarification required before taking action.
- Review — The human confirms whether the action should be taken.
“It’s not deterministic,” added Jokel. “It doesn’t always give you the same outcome, and we can build scaffolding, but ultimately you still ant a human being sitting at the keyboard checking to make sure that this decision is the right thing to do before it gets executed, and I think we’ll be in that state for a relatively long period of time.”
It is difficult to imagine how the same underlying technology that powers AI chatbots such as ChatGPT, which often hallucinate when asked simple prompts, would be able to power these ambient AI experiences.
Vijoy Pandey, GM and senior vice president of Outshift, said a major difference lies in the specialization of the models.
“Like in the movie-making industry, actors then come together and create a movie; it’s the same process,” said Pandey to ZDNET. “All of these are really specific models and agents that do specific tasks and not broad.”
Also: The best AI chatbots: ChatGPT, Copilot, and notable alternatives
You wouldn’t expect a human who is a subject matter expert to know everything about every topic, Pandey explained. Yet, people do expect that from tools like ChatGPT, which leads to AI hallucination. What’s needed are AI agents that are subject matter experts and are really good at a narrow task.
So, when can we expect this type of technology to emerge? Chase shared on stage that while it is still early days, it is “absolutely where we are headed.”
Want more stories about AI? Sign up for Innovation, our weekly newsletter.
